feat: audio support record (#29)

src/hooks/useSpeechService.ts

@@ -2,6 +2,7 @@ import type { VoiceInfo } from 'microsoft-cognitiveservices-speech-sdk'
 import {
   AudioConfig,
   CancellationErrorCode,
+  ResultReason,
   SpeakerAudioDestination,
   SpeechConfig,
   SpeechRecognizer,
@@ -52,6 +53,10 @@ export const useSpeechService = ({ langs = <const>['fr-FR', 'ja-JP', 'en-US', 'z
   // const isFetchAllVoices = ref(false) // 是否在请求所有语音列表
   const rate = ref(1) // 语速 (0,2]
 
+  let mediaRecorder: MediaRecorder | null
+  const chunks: Blob[] = []
+  const audioBlob = ref<Blob>(new Blob())
+
   const allVoices = ref<VoiceInfo[]>([])
 
   const recognizer = ref<SpeechRecognizer>(new SpeechRecognizer(speechConfig.value))
@@ -89,13 +94,29 @@ export const useSpeechService = ({ langs = <const>['fr-FR', 'ja-JP', 'en-US', 'z
     immediate: true,
   })
 
-  // watch([azureKey, azureRegion], () => {
-  //   if (isFetchAllVoice && allVoices.value.length === 0)
-  //     getVoices()
-  // })
-
   // 语音识别
-  const startRecognizeSpeech = (cb?: (text: string) => void) => {
+
+  const audioRecorder = async () => {
+    // 暂时通过 mediaRecorder 方式实现录音保存，后续可能会改为直接通过 SpeechRecognizer 实现保存
+
+    const stream = await navigator.mediaDevices.getUserMedia({ audio: true })
+    mediaRecorder = new MediaRecorder(stream)
+
+    mediaRecorder.ondataavailable = (e) => {
+      chunks.push(e.data)
+    }
+
+    mediaRecorder.onstop = (e) => {
+      const blob = new Blob(chunks, { type: 'audio/wav' })
+      audioBlob.value = blob
+      mediaRecorder = null
+      chunks.length = 0
+    }
+
+    mediaRecorder.start()
+  }
+
+  const startRecognizeSpeech = async (cb?: (text: string) => void) => {
     isRecognizReadying.value = true
 
     recognizer.value.canceled = () => {
@@ -105,9 +126,10 @@ export const useSpeechService = ({ langs = <const>['fr-FR', 'ja-JP', 'en-US', 'z
       console.log('Recognize result: ', e.result.text)
       cb && cb(e.result.text)
     }
-    recognizer.value.recognizing = (s, e) => {
-      console.log('Recognize recognizing', e.result.text)
+    recognizer.value.recognizing = (s, event) => {
+      console.log('Recognize recognizing', event.result.text)
     }
+
     recognizer.value.sessionStopped = (s, e) => {
       console.log('\n    Session stopped event.')
       recognizer.value.stopContinuousRecognitionAsync()
@@ -122,7 +144,8 @@ export const useSpeechService = ({ langs = <const>['fr-FR', 'ja-JP', 'en-US', 'z
       isRecognizing.value = false
     }
 
-    recognizer.value.startContinuousRecognitionAsync(() => {
+    recognizer.value.startContinuousRecognitionAsync(async () => {
+      await audioRecorder()
       isRecognizing.value = true
       isRecognizReadying.value = false
       console.log('Recognize...')
@@ -137,6 +160,8 @@ export const useSpeechService = ({ langs = <const>['fr-FR', 'ja-JP', 'en-US', 'z
 
   // 停止语音识别
   const stopRecognizeSpeech = (): Promise<void> => {
+    mediaRecorder!.stop()
+
     isRecognizReadying.value = false
     return new Promise((resolve, reject) => {
       recognizer.value.stopContinuousRecognitionAsync(() => {
@@ -267,5 +292,6 @@ export const useSpeechService = ({ langs = <const>['fr-FR', 'ja-JP', 'en-US', 'z
     allVoices,
     isSynthesizing,
     rate,
+    audioBlob,
   }
 }

src/pages/Home/components/Content.vue

@@ -1,7 +1,7 @@
 <script setup lang="ts">
 import Button from '@/components/Button.vue'
 import { generatTranslate, generateText } from '@/server/api'
-import { verifyOpenKey } from '@/utils'
+import { base64ToBlob, blobToBase64, verifyOpenKey } from '@/utils'
 import { useConversationStore } from '@/stores'
 
 interface Translates {
@@ -27,6 +27,7 @@ const {
   stopRecognizeSpeech,
   ssmlToSpeak,
   isSynthesizing,
+  audioBlob,
 } = useSpeechService({ langs: store.allLanguage as any, isFetchAllVoice: false })
 
 // states
@@ -106,10 +107,11 @@ async function onSubmit() {
 
   store.changeConversations([
     ...currentChatMessages.value,
-    { content: message.value, role: 'user' },
+    { content: message.value, role: 'user', audioBlob: await blobToBase64(audioBlob.value) },
   ])
+  const tempCurrentChatMessages = currentChatMessages.value.map(x => ({ content: x.content, role: x.role })) // 发送的请求中需去除audioBlob
   const systemMessage = currentChatMessages.value[0]
-  const relativeMessage = [...chatMessages.value, { content: message.value, role: 'user' }].slice(-(Number(chatRememberCount.value))) // 保留最近的几条消息
+  const relativeMessage = [...tempCurrentChatMessages, { content: message.value, role: 'user' }].slice(-(Number(chatRememberCount.value))) // 保留最近的几条消息
   const prompts = [systemMessage, ...relativeMessage] as ChatMessage[]
 
   message.value = ''
@@ -134,13 +136,40 @@ async function onSubmit() {
   store.changeLoading(false)
 }
 
+// assistant speak
 function speak(content: string, index: number) {
+  restartAudio()
   if (isPlaying.value || isSynthesizing.value) return
   speakIndex.value = index
   text.value = content
   ssmlToSpeak(content)
 }
 
+// user speak
+let audio = new Audio()
+
+function restartAudio() {
+  audio.pause()
+  audio.currentTime = 0
+  isPlaying.value = false
+  // audio.play()
+}
+
+function userSpeak(audioData: string, index: number) {
+  if (isPlaying.value || isSynthesizing.value) return
+  speakIndex.value = index
+  audio = new Audio(URL.createObjectURL(base64ToBlob(audioData)))
+  audio.play()
+  audio.onplay = () => {
+    isPlaying.value = true
+  }
+
+  audio.onended = () => {
+    isPlaying.value = false
+    speakIndex.value = -1
+  }
+}
+
 const recognize = async () => {
   try {
     console.log('isRecognizing', isRecognizing.value)
@@ -194,15 +223,15 @@ const translate = async (text: string, i: number) => {
           <div class="w-10 h-10">
             <img w-full h-full object-fill rounded-full :src="item.role === 'user' ? selfAvatar : currentAvatar" alt="">
           </div>
-
           <div style="flex-basis:fit-content" mx-2>
             <p p-2 my-2 chat-box>
               {{ item.content }}
             </p>
-            <p v-show="item.role === 'assistant' && translates[item.content + i]?.isShow " p-2 my-2 chat-box>
+            <p v-show=" translates[item.content + i]?.isShow " p-2 my-2 chat-box>
               {{ translates[item.content + i]?.result }}
             </p>
 
+            <!-- assistant -->
             <p v-if="item.role === 'assistant'" mt-2 flex>
               <template v-if="speakIndex !== i">
                 <span class="chat-btn" @click="speak(item.content, i)">
@@ -224,6 +253,31 @@ const translate = async (text: string, i: number) => {
                 <i icon-btn i-eos-icons:bubble-loading />
               </span>
             </p>
+
+            <!-- user -->
+            <p v-else mt-2 flex>
+              <template v-if="item.audioBlob">
+                <template v-if="speakIndex !== i">
+                  <span class="chat-btn" @click="userSpeak(item.audioBlob, i)">
+                    <i icon-btn rotate-270 i-ic:sharp-wifi />
+                  </span>
+                </template>
+                <template v-else>
+                  <span v-if="isPlaying" class="chat-btn" @click="restartAudio()">
+                    <i icon-btn rotate-270 i-svg-spinners:wifi-fade />
+                  </span>
+                  <span v-else class="chat-btn" @click="userSpeak(item.audioBlob, i)">
+                    <i icon-btn rotate-270 i-ic:sharp-wifi />
+                  </span>
+                </template>
+              </template>
+              <span v-if="!isTranslating || translateIndex !== i" ml-1 class="chat-btn" @click="translate(item.content, i)">
+                <i icon-btn i-carbon:ibm-watson-language-translator />
+              </span>
+              <span v-else ml-1 class="chat-btn">
+                <i icon-btn i-eos-icons:bubble-loading />
+              </span>
+            </p>
           </div>
         </div>
       </template>

src/stores/index.ts

@@ -24,7 +24,7 @@ export interface Conversation {
   key: Key // 名称 唯一标识
   name: string // 名称
   desc: string
-  chatMessages: ChatMessage[] // 聊天信息
+  chatMessages: ChatMessageWithAudioUrl[] // 聊天信息
   language: string // tts stt
   voice: string // 参考 https://aka.ms/speech/tts-languages
   avatar: string // 用户头像
@@ -66,7 +66,7 @@ export const useConversationStore = defineStore('conversation', {
     },
   },
   actions: {
-    changeConversations(chatMessages: ChatMessage[]) {
+    changeConversations(chatMessages: ChatMessageWithAudioUrl[]) {
       this.chatMessages(this.currentKey)!.chatMessages = chatMessages
     },
     changeCurrentKey(key: Key) {

src/types..d.ts

@@ -27,3 +27,7 @@ interface ImagePayload {
   n?: number
   size?: string
 }
+
+interface ChatMessageWithAudioUrl extends ChatMessage {
+  audioBlob?: string //base64
+}

src/utils/tools.ts

@@ -25,3 +25,29 @@ export async function fetchWithTimeout(
 
   return response
 }
+
+export function blobToBase64(blob: Blob) {
+  return new Promise<string>((resolve, reject) => {
+    const reader = new FileReader()
+    if (blob.size === 0)
+      return resolve('')
+
+    reader.readAsDataURL(blob)
+    reader.onload = function () {
+      const dataUrl = reader.result
+      resolve(dataUrl!.toString())
+    }
+    reader.onerror = reject
+  })
+}
+
+export function base64ToBlob(dataUrl: string) {
+  const arr = dataUrl.split(',')
+  const mime = arr[0].match(/:(.*?);/)![1]
+  const bstr = atob(arr[1])
+  let n = bstr.length
+  const u8arr = new Uint8Array(n)
+  while (n--)
+    u8arr[n] = bstr.charCodeAt(n)
+  return new Blob([u8arr], { type: mime })
+}